Legal claims defining the scope of protection. Each claim is shown in both the original legal language and a plain English translation.
1. A computer-implemented method of segmenting and ranking nuclei in source images of a tissue specimen, the tissue specimen stained with at least a first and a second in situ hybridization stain having different colors comprising (1) computing a foreground segmentation mask based on an input image of the tissue specimen; (2) identifying individual nuclei by filtering the input image with the computed foreground segmentation mask; (3) computing metrics for all identified nuclei based on black in situ hybridization signals corresponding to HER2 and red in situ hybridization signals corresponding to Chromosome 17 present in the nuclei; (4) evaluating the metrics to determine nuclei suitable for ranking, wherein nuclei suitable for ranking have at least one black dot corresponding to black in situ hybridization signals, at least one red dot corresponding to red in situ hybridization signals, and satisfy a predetermined area constraint; and (5) identifying a top N number of nuclei for scoring from the determined nuclei suitable for ranking, wherein the top N number of nuclei for scoring are identified by (1) identifying a total number of nuclei that meet established criteria; (2) segregating the identified total number of nuclei into three subsets, where a first subset comprises nuclei ranked as having the highest average absorbance, a second subset comprises nuclei ranked as having the highest average A, and a third subset comprises nuclei common to both the first and second subsets; and (3) evaluating whether a number of nuclei in the third subset meets a predefined threshold amount.
This invention relates to a computer-implemented method for segmenting and ranking nuclei in tissue specimen images stained with two in situ hybridization (ISH) stains of different colors, such as black for HER2 and red for Chromosome 17. The method addresses the challenge of accurately identifying and ranking nuclei based on their ISH signal characteristics, which is critical for diagnostic applications like HER2 testing in breast cancer. The method begins by computing a foreground segmentation mask from an input image of the stained tissue specimen to isolate the relevant tissue area. Individual nuclei are then identified by applying this mask to filter the input image. For each identified nucleus, metrics are computed based on the presence and intensity of black (HER2) and red (Chromosome 17) ISH signals. These metrics are evaluated to determine which nuclei are suitable for ranking, requiring at least one black dot, at least one red dot, and adherence to a predetermined area constraint. The suitable nuclei are then ranked to identify the top N nuclei for scoring. This involves calculating the total number of nuclei meeting the criteria and segregating them into three subsets: those with the highest average absorbance, those with the highest average A (likely a specific metric), and those common to both subsets. The method checks if the number of nuclei in the overlapping subset meets a predefined threshold, ensuring reliable scoring. This approach improves the accuracy and efficiency of nuclear analysis in diagnostic pathology.
2. The computer-implemented method of claim 1 , wherein the foreground segmentation mask is computed by (1) applying filters to enhance the input image such that (a) image regions unlikely to have nuclei are discarded, and (b) nuclei within a local region are identified; and (2) further applying optional filters to selectively remove artifacts, remove small blobs, remove small discontinuities, fill holes, and split up bigger blobs, wherein the filters applied are selected from a group consisting of a global thresholding filter, a locally adaptive thresholding filter, morphological operations, and watershed transformations.
This invention relates to image processing techniques for identifying and segmenting nuclei in biological images. The problem addressed is the accurate and efficient segmentation of nuclei from complex biological images, which often contain artifacts, varying lighting conditions, and overlapping structures. The method enhances the input image by applying filters to discard regions unlikely to contain nuclei and to identify nuclei within local regions. This involves initial filtering to suppress non-nuclei areas and highlight potential nuclei. Additional optional filters are then applied to refine the segmentation, including removing artifacts, small blobs, and discontinuities, filling holes, and splitting larger blobs. The filters used may include global or locally adaptive thresholding, morphological operations, and watershed transformations. These steps collectively improve the accuracy of nucleus detection and segmentation in biological imaging applications. The method is designed to handle challenges such as noise, varying contrast, and overlapping structures, ensuring reliable nucleus identification for further analysis.
3. The computer-implemented method of claim 2 , wherein the global thresholding filter is applied first, followed by application of the locally adaptive thresholding filter.
4. The computer-implemented method of claim 1 , wherein the identification of the individual nuclei further comprises performing a connected components labeling process on the filtered input image.
5. The computer-implemented method of claim 1 , wherein the metrics based on the black and red in situ hybridization signals present in the identified nuclei are selected from a group consisting of an average absorbance metric and an average A channel metric, wherein the average absorbance metric is computed within a RGB domain at all local maxima using radial symmetry voting on gradient magnitude of difference of Gaussian applied on an absorbance channel, and wherein the average A channel metric is computed within a LAB color space at all local maxima using radial symmetry voting on gradient magnitude of difference of Gaussian applied on an A channel.
6. The computer-implemented method of claim 1 , wherein the number of nuclei within the third subset are ranked based on a black-to-red ratio.
7. A non-transitory computer-readable medium for storing computer-executable instructions that are executed by a processor to perform operations comprising: (1) computing a foreground segmentation mask based on an input image; (2) identifying individual nuclei by filtering the input image with the computed foreground segmentation mask; (3) computing metrics for all identified nuclei based on black in situ hybridization signals corresponding to HER2 and red in situ hybridization signals corresponding to Chromosome 17; (4) determining nuclei suitable for ranking based on the computed metrics, wherein the nuclei suitable for ranking have at least one black dot corresponding to black in situ hybridization signals, at least one red dot corresponding to red in situ hybridization signals, and satisfy a predetermined area constraint; and (5) identify a top N number of nuclei for scoring, and scoring the identified nuclei, wherein the top N number of nuclei for scoring are identified by (1) identifying a total number of nuclei that meet established criteria; (2) segregating the identified total number of nuclei into three subsets, where a first subset comprises nuclei ranked as having the highest average absorbance, a second subset comprises nuclei ranked as having the highest average A, and a third subset comprises nuclei common to both the first and second subsets; and (3) evaluating whether a number of nuclei in the third subset meets a predefined threshold amount, and wherein the number of nuclei within the third subset are ranked based on a black to red ratio.
8. The non-transitory computer-readable medium of claim 7 , wherein the foreground segmentation mask is computed by (1) applying filters to enhance the input image such that (a) image regions unlikely to have nuclei are discarded, and (b) nuclei within a local region are identified; and (2) further applying optional filters to selectively remove artifacts, remove small blobs, remove small discontinuities, fill holes, and split up bigger blobs, wherein the filters applied are selected from a group consisting of a global thresholding filter, a locally adaptive thresholding filter, morphological operations, and watershed transformations, and wherein the global thresholding filter is applied first, followed by application of the locally adaptive thresholding filter.
9. The non-transitory computer-readable medium of claim 8 , wherein the optional filters to selectively remove artifacts, remove small blobs, remove small discontinuities, fill holes, and split up bigger blobs are applied after application of the locally adaptive thresholding filter.
This invention relates to image processing techniques for enhancing binary images, particularly in medical imaging applications. The problem addressed is the presence of artifacts, small blobs, discontinuities, and other imperfections in binary images that degrade their quality and usability. The solution involves a sequence of filtering operations applied to a binary image to improve its structural integrity and readability. The process begins with a locally adaptive thresholding filter, which converts a grayscale image into a binary image by adaptively adjusting threshold values based on local image characteristics. This step helps in distinguishing foreground objects from the background more accurately. Following this, optional filters are applied to further refine the binary image. These filters include artifact removal to eliminate unwanted noise, small blob removal to clean up isolated pixels or small groups of pixels, discontinuity removal to smooth out jagged edges, hole filling to close small gaps within objects, and blob splitting to separate merged objects. The order of operations ensures that the locally adaptive thresholding provides a clean base image before applying these additional refinements. The invention is particularly useful in medical imaging, where precise and clear binary images are essential for diagnosis and analysis. The combination of adaptive thresholding followed by targeted filtering improves the accuracy and reliability of image-based medical assessments.
10. The non-transitory computer-readable medium of claim 9 , wherein the identification of individual nuclei comprises performing a connected components labeling process on the filtered input image.
11. The non-transitory computer-readable medium of claim 10 , wherein the metrics based on black and red in situ hybridization signals present in the identified nuclei are selected from a group consisting of an average absorbance metric and an average A channel metric, wherein the average absorbance metric is computed within a RGB domain at all local maxima using radial symmetry voting on gradient magnitude of difference of Gaussian applied on an absorbance channel, and wherein the average A channel metric is computed within a LAB color space at all local maxima using radial symmetry voting on gradient magnitude of difference of Gaussian applied on an A channel.
Unknown
February 2, 2021
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.